One key.
Every model. Anywhere.
A working gateway to Claude, GPT, Gemini, DeepSeek, Kimi and ninety more models — kept alive by a community pool of keys that rotate themselves when one burns out.
curl https://dizosrf-firepass.hf.space/v1/chat/completions \
-H "Authorization: Bearer sk-firepass-…" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-opus-4-7",
"messages": [{"role":"user","content":"hello"}]
}'Built to just work
Self-healing pool
A community-sourced pool of upstream keys is checked every minute. The dead ones are quarantined automatically; the next request walks on to the next healthy key.
Drop-in APIs
/v1/chat/completions, /v1/models, and a fully-translated Anthropic /anthropic/v1/messages. The official SDKs work unchanged.
Streaming
Server-sent events for both OpenAI and Anthropic shapes. Low TTFB, no buffering, no clever shortcuts — just the bytes you asked for, in order.
Your own key
Sign up, mint a personal sk-firepass-…, pin the rate limit you want, and revoke it any time. No surprise bills, no long contracts.
Smart aliases
Ask for best, fast, or cheap — the gateway picks a real model that fits. Pin a specific model when you need to be precise.
Open upstream
Keys are imported from a public, community-maintained list and refreshed whenever it changes. Always transparent, always current — never a black box.
Three steps, then you're in
You sign up
Email + password + captcha. We email a 6-digit code. You verify. You pick a key name and a rate limit.
You point your tool
opencode, Claude Code, Cursor, ChatBox, the openai SDK — anything that talks OpenAI or Anthropic. Set the base URL, paste the key, ship.
We keep the fire lit
Every minute: poll upstream, score each key, retire the broken ones, queue the fresh ones. You see fewer 429s and more 200s.